In April 2026, a group of international human rights experts and organizations gathered in Gilon, Switzerland, to grapple with a difficult question: How can governments and technology companies address concerns about the spread of incitement to hatred on social media without running afoul of international human rights standards that protect freedom of expression?
The meeting included ARTICLE 19, the Swiss Federal Department of Foreign Affairs and the Office of the United Nations High Commissioner for Human Rights, and the result was a note directed to the Office of the United Nations High Commissioner for Human Rights titled “The Glion Recommendations on Implementing the Rabat Plan of Action in the Digital Age.”
“The rapid expansion of social media, alongside the use of artificial intelligence and other technologies,” they wrote, “has fundamentally altered the scale, speed and reach of expression and its impact.” They warn that “speech that may amount to incitement” can now “be disseminated and amplified rapidly both within and across borders and jurisdictions.”
While safeguarding freedom of expression is also a stated goal of the initiative, the stakeholders — who express serious concerns about the spread of online hate speech — are grappling with a longstanding tension in international human rights law: where to draw the line between protecting freedom of expression and prohibiting incitement to hatred.
This tension arises from two provisions in the International Covenant on Civil and Political Rights (ICCPR). While Article 19 protects freedom of expression, Article 20(2) requires states to prohibit “any advocacy of national, racial or religious hatred that constitutes incitement to discrimination, hostility or violence.”
In 2012, the United Nations (UN) General Assembly adopted the Rabat Plan of Action (RPA/Rabat), which created a six-part threshold test to identify expression that constitutes a criminal offense under Article 20(2). The test requires states to consider factors such as context, speaker, intent, content and form, extent of the speech act, and the likelihood of harm, including imminence.
Since Rabat’s adoption, the Gilon Recommendations rightly point out, the challenge of addressing incitement to hatred while safeguarding freedom of expression has taken on new dimensions. The new expressive reality of social media, algorithms, and AI has dramatically changed how speech develops and spreads. The recommendations attempt to offer non-exhaustive guidance to states, private companies, and other stakeholders, such as the UN and other NGOs, on how to apply the Rabat threshold test to the online sphere.
But transposing strict human rights standards into digital governance is an intricately complex task. Can RPA’s strict threshold tests stand up to the pressures of our current digital reality, marked by phenomena like mass content moderation? Do the recommendations effectively address issues like a speaker’s intent in an online environment, especially when algorithmic amplification and other factors outside a speaker’s control are at play?
This essay argues that the Gilon Recommendations fall short on these and other questions. Beyond the fact that international human rights law’s prohibition on certain forms of incitement inherently limits free expression, empirical evidence that broader hate speech restrictions effectively reduce hatred, discrimination, or violence remains limited and contested — factors also omitted from this analysis.
Rabat and Glion
Rabat’s six-part test was developed to assess when incitement is sufficiently serious to warrant criminal sanctions. Glion seeks to extend the RPA’s application to the wider regulation of online expression, including platform content removal.
Criminal punishment differs markedly from content moderation, with different consequences and, therefore, different thresholds. Nevertheless, the Glion variation of the Rabat thresholds creates significant complexity, as the recommendations recognize that online expression may not meet incitement thresholds but may still cause harm because of its methods of dissemination and amplified reach in the digital ecosystem.
While this is understandable in practice, the legal question is whether a framework developed to ensure that the criminalization of speech is a ‘last resort’ can be adapted to wider forms of regulation without undermining those protections. By seeking to transpose Rabat broadly to the online sphere, there is a risk that Rabat’s authority could be used to justify restrictions on expression that its original test was designed to protect. Given the current deterioration of the protection of free speech, particularly in the digital framework, these risks could not be higher.
Moreover, the application of Rabat to the digital age poses a more fundamental question — beyond the fact that this may also result in the restriction of ‘lawful but awful speech’ — that the recommendation fails to address: Can measures designed to combat hatred also restrict expression without necessarily reducing the harms they are meant to prevent?
The existence of an obligation to prohibit some forms of incitement does not mean that broader hate speech restrictions are effective. Indeed, empirical research suggests reasons to question this assumption. A study about the prosecution of Dutch MP Geert Wilders for hate speech found that the judgment against him positively correlated with an increase in support for his party amongst certain categories of voters. In other words, speech restrictions imposed on a controversial political actor can actually increase their support and popularity..
Further, high levels of free speech protections are correlated with lower levels of social conflict, suggesting that more speech-restrictive measures may create resentment, drive extremism underground, and cut off the safety-valve effect that free expression provides to individuals and groups with grievances.
Against this backdrop, the Glion Recommendations must pass a strict scrutiny test to ensure they remain compatible with international human rights law safeguards and do not produce unintended consequences from lowering Rabat thresholds.
Foreseeability and The Weakening of Intent
One of Glion’s most problematic provisions relates to the interpretation of intent. The Recommendations suggest that, when assessing whether a speaker intended to incite hatred, consideration should be given to whether the speaker could have foreseen a serious risk of discrimination, hostility or violence.
Although foreseeability may provide relevant evidence, it should not be confused with intention. Rabat is explicit that Article 20(2) requires intent and that negligence and recklessness are insufficient for criminal liability. Specifically, Article 29(1) provides that “Article 20 of the International Covenant on Civil and Political Rights anticipates intent. Negligence and recklessness are not sufficient for an act to be an offense under Article 20 of the Covenant, as this article provides for ‘advocacy’ and ‘incitement’ rather than the mere distribution or circulation of material.”
Consider a women’s rights activist who criticizes discriminatory practices within a religious community, knowing that her comments may fuel hostility towards its members. Does such a risk constitute an intention to incite hate? If the answer is yes, can criticism of discrimination potentially be restricted because of how communities and groups may react to it? If so, this would contravene international standards.
For example, the 2001 Joint Statement on Racism and the Media — signed by the respective freedom of expression rapporteurs/representatives — expressly cautioned against penalizing the dissemination of hate speech without an intention to incite discrimination, hostility or violence. Similarly, the UN Special Rapporteur on Freedom of Expression defined advocacy under Article 20(2) ICCPR as “explicit, intentional, public and active support and promotion of hatred towards the target group.”
Glion’s appendix also proposes considering elements such as a speaker’s posting history, financial incentives, and the use of bots and automated accounts to amplify the speech. However, it does not explain how to balance these factors within the broader environment of freedom of expression, nor does it determine whether any factor alone could prove intent. Absent more specific guidance, the above factors alone or in conjunction with each other can override a speaker’s intent.
Another issue is Glion’s proposals that the context of cross-border speech is primarily assessed through the receiving audience’s context. But the question of whether a speaker in location X can duly comprehend the impact and consequences of their speech on a listener in location Y, with all its social, historical and/or political particularities, creates an insurmountable burden on the speaker and would no doubt consider numerous unintended effects outside of a speaker’s control.
The issue of assessing context, not only geographically but also in terms of a particular community, is further complicated when automated moderation systems are used. As Thiago Dias Oliva observes, such systems “still fail to understand context, thereby posing risks to users’ free speech, access to information and equality.”
Algorithmic Amplification
One of Glion’s important contributions is recognizing that the risks associated with online incitement cannot be understood independently of platform architecture, including engagement-driven business models and recommendation systems. As such, Glion recommends that companies mitigate risks associated with content curation, monetization, and algorithmic amplification.
However, responsibility for creating potentially inciting speech is not necessarily equivalent to responsibility for amplifying it. Glion notes that “where bots or automated accounts are used to amplify the speaker’s content, treat the network as an extension of the speaker’s reach.” This could be considered reasonable where a speaker deliberately coordinates amplification. But it becomes even more complex when third parties disseminate the speech, affecting listeners who were not originally intended as the speaker’s target audience.
As Terleton Gillespie argues, algorithmic systems are not neutral conduits of information; they can shape the reach and effects of speech. So, saying that enhanced dissemination equates to enhanced impact cannot, under international standards, meet intention thresholds. As such, clarity is needed on attribution. Otherwise, assessing whether the threshold of incitement has been met risks becoming mired in technological issues over which speakers have limited control.
From Principles to Accountability
The Glion Recommendations mark a significant development in keeping international human rights standards alive in the digital world. Yet its over-broad approach raises a difficult question: how can international standards constrain private regulatory power without inadvertently legitimizing excessive restrictions on lawful speech?
As legal expert Nicolas Suzor argues, platform governance can be legitimate if it depends on rule of law principles. Evelyn Douek similarly challenges the adequacy of individualized remedies, emphasizing the need to scrutinize the systems through which moderation decisions are produced. These insights suggest that implementing Glion requires more than incorporating human rights language into corporate policies.
Moreover, as the empirical evidence mentioned above demonstrates, hate speech laws do not actually result in a fall in hate speech. In fact, in many instances, they do backfire. The Recommendations do not adequately address this issue, leaving an empirical gap in the final output.
Adapting such a narrowly conceived framework to this wider context without diluting its safeguards is inherently challenging, and getting it wrong has significant consequences. Ultimately, the question is whether the restrictions Glion may encourage are necessary, proportionate and demonstrably effective. Without convincing answers, extending Rabat’s framework risks turning a safeguard against excessive interference with freedom of expression into a justification for precisely that interference.
Natalie Alkiviadou is a Senior Research Fellow at The Future of Free Speech. Her research interests lie in the freedom of expression, the far-right, hate speech, hate crime, and non-discrimination.



